FrontierSeptember 18, 2026via MarkTechPost
PrismML Releases Ternary Bonsai 2 27B: A 5.9 GB Apache 2.0 Model Retaining 98.2% of Qwen3.8 27B Performance
Why it matters
Quantization breakthrough that dramatically shrinks a capable open-weight model while retaining near-parity performance—matters for practitioners deploying locally or on-device, and signals the efficiency frontier is moving faster than capability.
Key signals
- Ternary Bonsai 2 27B: 5.93 GB (vs 53.80 GB FP16)
- Retains 98.2% of parent model performance across 20 benchmarks
- Multimodal: text + images, 262K token context
- Apache 2.0 license (open-weight)
- Deployed in Cline coding agent demo
- ternary-weight quantization approach
- Ternary Bonsai 2 27B: 5.93 GB vs 53.80 GB FP16 original (~89% size reduction)
- Based on Qwen3.8 27B, multimodal (text + images)
- 262K-token context window
- Apache 2.0 license (open-weight, commercial-friendly)
- Demonstrated running Cline coding agent
- MarkTechPost, published Sep 18, 2026
The hook
98.2% performance in 5.9 GB. PrismML's ternary-weight drop makes Qwen3.8 27B deployable on edge and consumer hardware.
PrismML has released Ternary Bonsai 2 27B, a ternary-weight version of Qwen3.8 27B. The language model occupies 5.93 GB, against 53.80 GB in FP16. PrismML reports that it keeps 98.2% of the parent model’s average across 20 benchmarks. The model accepts text and images and supports a 262K-token conte…